Total least squares

In the least squares method of data modeling, the objective function to be minimized, S, is a quadratic form:

:$S=\backslash mathbf\{r^TWr\},$

where r is the vector of residuals and W is a weighting matrix. In linear least squares the model contains equations which are linear in the parameters appearing in the parameter vector $\backslash boldsymbol\backslash beta$, so the residuals are given by

:$\backslash mathbf\{r=y-X\backslash boldsymbol\backslash beta\}.$

There are m observations in y and n parameters in β with m>n. X is a m×n matrix whose elements are either constants or functions of the independent variables, x. The weight matrix W is, ideally, the inverse of the variance-covariance matrix $\backslash mathbf\; M\_y$ of the observations y. The independent variables are assumed to be error-free. The parameter estimates are found by setting the gradient equations to zero, which results in the normal equations

An alternative form is $\backslash mathbf\{X^TWX\backslash boldsymbol\backslash Delta\; \backslash boldsymbol\backslash beta=X^T\; W\; \backslash boldsymbol\backslash Delta\; y\}$, where $\backslash boldsymbol\backslash Delta\; \backslash boldsymbol\backslash beta$ is the parameter shift from some starting estimate of $\backslash boldsymbol\backslash beta$ and $\backslash boldsymbol\backslash Delta\; \backslash mathbf\; y$ is the difference between y and the value calculated using the starting value of $\backslash boldsymbol\backslash beta$

:$\backslash mathbf\{X^TWX\backslash boldsymbol\backslash beta=X^T\; Wy\}.$

Now, suppose that both x and y are observed subject to error, with variance-covariance matrices $\backslash mathbf\; M\_x$ and $\backslash mathbf\; M\_y$ respectively. In this case the objective function can be written as

:$S=\backslash mathbf\{r\_x^TM\_x^\{-1\}r\_x+r\_y^TM\_y^\{-1\}r\_y\},$

where $\backslash mathbf\; r\_x$ and $\backslash mathbf\; r\_y$ are the residuals in x and y respectively. Clearly{{Explain|date=February 2021}} these residuals cannot be independent of each other, but they must be constrained by some kind of relationship. Writing the model function as $\backslash mathbf\{f(r\_x,r\_y,\backslash boldsymbol\backslash beta)\}$, the constraints are expressed by m condition equations.W.E. Deming, Statistical Adjustment of Data, Wiley, 1943

:$\backslash mathbf\{F=\backslash Delta\; y\; -\backslash frac\{\backslash partial\; f\}\{\backslash partial\; r\_x\}\; r\_x-\backslash frac\{\backslash partial\; f\}\{\backslash partial\; r\_y\}\; r\_y\; -X\backslash Delta\backslash boldsymbol\backslash beta=0\}.$

Thus, the problem is to minimize the objective function subject to the m constraints. It is solved by the use of Lagrange multipliers. After some algebraic manipulations,{{cite book |last=Gans |first=Peter |title=Data Fitting in the Chemical Sciences |year=1992 |publisher=Wiley |isbn=9780471934127 |url=http://www.wiley.com/WileyCDA/WileyTitle/productCd-0471934127.html |access-date=4 December 2012}} the result is obtained.

:$\backslash mathbf\{X^TM^\{-1\}X\backslash Delta\; \backslash boldsymbol\backslash beta=X^T\; M^\{-1\}\; \backslash Delta\; y\},$

or alternatively $\backslash mathbf\{X^TM^\{-1\}X\; \backslash boldsymbol\backslash beta=X^T\; M^\{-1\}\; y\},$

where M is the variance-covariance matrix relative to both independent and dependent variables.

:$\backslash mathbf\{M=K\_xM\_xK\_x^T+K\_yM\_yK\_y^T;\backslash \; K\_x=-\backslash frac\{\backslash partial\; f\}\{\backslash partial\; r\_x\},\backslash \; K\_y=-\backslash frac\{\backslash partial\; f\}\{\backslash partial\; r\_y\}\}.$

When the data errors are uncorrelated, all matrices M and W are diagonal. Then, take the example of straight line fitting.

:$f(x\_i,\backslash beta)=\backslash alpha\; +\; \backslash beta\; x\_i$

in this case

:$M\_\{ii\}=\backslash sigma^2\_\{y,i\}+\backslash beta^2\; \backslash sigma^2\_\{x,i\}$

showing how the variance at the ith point is determined by the variances of both independent and dependent variables and by the model being used to fit the data. The expression may be generalized by noting that the parameter $\backslash beta$ is the slope of the line.

:$M\_\{ii\}=\backslash sigma^2\_\{y,i\}+\backslash left(\backslash frac\{dy\}\{dx\}\backslash right)^2\_i\; \backslash sigma^2\_\{x,i\}$

An expression of this type is used in fitting pH titration data where a small error on x translates to a large error on y when the slope is large.

As was shown in 1980 by Golub and Van Loan, the TLS problem does not have a solution in general.G. H. Golub and C. F. Van Loan, An analysis of the total least squares problem. Numer. Anal., 17, 1980, pp. 883–893. The following considers the simple case where a unique solution exists without making any particular assumptions.

The computation of the TLS using singular value decomposition (SVD) is described in standard texts.{{Cite book

| last1=Golub |first1=Gene H. |author-link1=Gene H. Golub

| last2=Van Loan |first2=Charles F. |author-link2=Charles F. Van Loan

| title = Matrix Computations

| edition = 3rd

| publisher = The Johns Hopkins University Press

| year = 1996

}} pp 596. We can solve the equation

:$XB\; \backslash approx\; Y$

for B where X is m-by-n and Y is m-by-k. The notation XB ≈ Y is used here to reflect the notation used in the earlier part of the article. In the computational literature the problem has been more commonly presented as AX ≈ B, i.e. with the letter X used for the n-by-k matrix of unknown regression coefficients.

That is, we seek to find B that minimizes error matrices E and F for X and Y respectively. That is,

:$\backslash mathrm\{argmin\}\_\{B,E,F\}\; \backslash |\; [E\backslash ;\; F]\; \backslash |\_F,\; \backslash qquad\; (X+E)\; B\; =\; Y+F$

where $[E\backslash ;\; F]$ is the augmented matrix with E and F side by side and $\backslash |\backslash cdot\backslash |\_F$ is the Frobenius norm, the square root of the sum of the squares of all entries in a matrix and so equivalently the square root of the sum of squares of the lengths of the rows or columns of the matrix.

This can be rewritten as

:$[(X+E)\; \backslash ;\; (Y+F)]\; \backslash begin\{bmatrix\}\; B\backslash \backslash \; -I\_k\backslash end\{bmatrix\}\; =\; 0.$

where $I\_k$ is the $k\backslash times\; k$ identity matrix.

The goal is then to find $[E\backslash ;\; F]$ that reduces the rank of $[X\backslash ;\; Y]$ by k. Define $[U]\; [\backslash Sigma]\; [V]^*$ to be the singular value decomposition of the augmented matrix $[X\backslash ;\; Y]$.

:

where V is partitioned into blocks corresponding to the shape of X and Y.

Using the Eckart–Young theorem, the approximation minimising the norm of the error is such that matrices $U$ and $V$ are unchanged, while the smallest $k$ singular values are replaced with zeroes. That is, we want

:

so by linearity,

:

We can then remove blocks from the U and Σ matrices, simplifying to

:$[E\backslash ;\; F]\; =\; -U\_Y\backslash Sigma\_Y\; \backslash begin\{bmatrix\}V\_\{XY\}\backslash \backslash V\_\{YY\}\backslash end\{bmatrix\}^*=\; -[X\backslash ;\; Y]\; \backslash begin\{bmatrix\}V\_\{XY\}\backslash \backslash V\_\{YY\}\backslash end\{bmatrix\}\backslash begin\{bmatrix\}V\_\{XY\}\backslash \backslash \; V\_\{YY\}\backslash end\{bmatrix\}^*.$

This provides E and F so that

:$[(X+E)\; \backslash ;\; (Y+F)]\; \backslash begin\{bmatrix\}V\_\{XY\}\backslash \backslash \; V\_\{YY\}\backslash end\{bmatrix\}\; =\; 0.$

Now if $V\_\{YY\}$ is nonsingular, which is not always the case (note that the behavior of TLS when $V\_\{YY\}$ is singular is not well understood yet), we can then right multiply both sides by $-V\_\{YY\}^\{-1\}$ to bring the bottom block of the right matrix to the negative identity, givingBjõrck, Ake (1996) Numerical Methods for Least Squares Problems, Society for Industrial and Applied Mathematics. {{ISBN|978-0898713602}} {{page needed|date=June 2012}}

: $[(X+E)\; \backslash ;\; (Y+F)]\; \backslash begin\{bmatrix\}\; -V\_\{XY\}\; V\_\{YY\}^\{-1\}\; \backslash \backslash \; -V\_\{YY\}\; V\_\{YY\}^\{-1\}\backslash end\{bmatrix\}\; =\; [(X+E)\; \backslash ;\; (Y+F)]\; \backslash begin\{bmatrix\}\; B\backslash \backslash \; -I\_k\backslash end\{bmatrix\}\; =\; 0\; ,$

and so

:$B=-V\_\{XY\}\; V\_\{YY\}^\{-1\}.$

A naive GNU Octave implementation of this is:

function B = tls(X, Y)

[m n] = size(X); % n is the width of X (X is m by n)

Z = [X Y]; % Z is X augmented with Y.

[U S V] = svd(Z, 0); % find the SVD of Z.

VXY = V(1:n, 1+n:end); % Take the block of V consisting of the first n rows and the n+1 to last column

VYY = V(1+n:end, 1+n:end); % Take the bottom-right block of V.

B = -VXY / VYY;

end

The way described above of solving the problem, which requires that the matrix $V\_\{YY\}$ is nonsingular, can be slightly extended by the so-called classical TLS algorithm.S. Van Huffel and J. Vandewalle (1991) The Total Least Squares Problems: Computational Aspects and Analysis. SIAM Publications, Philadelphia PA.

The standard implementation of classical TLS algorithm is available through [http://www.netlib.org/vanhuffel/index.html Netlib], see also.S. Van Huffel, Documented Fortran 77 programs of the extended classical total least squares algorithm, the partial singular value decomposition algorithm and the partial total least squares algorithm, Internal Report ESAT-KUL 88/1, ESAT Lab., Dept. of Electrical Engineering, Katholieke Universiteit Leuven, 1988.S. Van Huffel, The extended classical total least squares algorithm, J. Comput. Appl. Math., 25, pp. 111–119, 1989. All modern implementations based, for example, on solving a sequence of ordinary least squares problems, approximate the matrix $B$ (denoted $X$ in the literature), as introduced by Van Huffel and Vandewalle. It is worth noting, that this $B$ is, however, not the TLS solution in many cases.M. Plešinger, The Total Least Squares Problem and Reduction of Data in AX ≈ B. Doctoral Thesis, TU of Liberec and Institute of Computer Science, AS CR Prague, 2008. Ph.D. ThesisI. Hnětynková, M. Plešinger, D. M. Sima, Z. Strakoš, and S. Van Huffel, The total least squares problem in AX ≈ B. A new classification with the relationship to the classical works. SIMAX vol. 32 issue 3 (2011), pp. 748–770.

For non-linear systems similar reasoning shows that the normal equations for an iteration cycle can be written as

:$\backslash mathbf\{J^TM^\{-1\}J\backslash Delta\; \backslash boldsymbol\backslash beta=J^T\; M^\{-1\}\; \backslash Delta\; y\},$

where $\backslash mathbf\{J\}$ is the Jacobian matrix.

{{main|Curve fitting#Algebraic fit versus geometric fit for curves}}

{{further|Orthogonal regression}}

When the independent variable is error-free a residual represents the "vertical" distance between the observed data point and the fitted curve (or surface). In total least squares a residual represents the distance between a data point and the fitted curve measured along some direction. In fact, if both variables are measured in the same units and the errors on both variables are the same, then the residual represents the shortest distance between the data point and the fitted curve, that is, the residual vector is perpendicular to the tangent of the curve. For this reason, this type of regression is sometimes called two dimensional Euclidean regression (Stein, 1983){{cite journal |last=Stein |first=Yaakov J. |title=Two Dimensional Euclidean Regression |url =http://www.dspcsp.com/pubs/euclreg.pdf }} or orthogonal regression.

A serious difficulty arises if the variables are not measured in the same units. First consider measuring distance between a data point and the line: what are the measurement units for this distance? If we consider measuring distance based on Pythagoras' Theorem then it is clear that we shall be adding quantities measured in different units, which is meaningless. Secondly, if we rescale one of the variables e.g., measure in grams rather than kilograms, then we shall end up with different results (a different line). To avoid these problems it is sometimes suggested that we convert to dimensionless variables—this may be called normalization or standardization. However, there are various ways of doing this, and these lead to fitted models which are not equivalent to each other. One approach is to normalize by known (or estimated) measurement precision thereby minimizing the Mahalanobis distance from the points to the line, providing a maximum-likelihood solution;{{Citation needed|date=July 2009}} the unknown precisions could be found via analysis of variance.

In short, total least squares does not have the property of units-invariance—i.e. it is not scale invariant. For a meaningful model we require this property to hold. A way forward is to realise that residuals (distances) measured in different units can be combined if multiplication is used instead of addition. Consider fitting a line: for each data point the product of the vertical and horizontal residuals equals twice the area of the triangle formed by the residual lines and the fitted line. We choose the line which minimizes the sum of these areas. Nobel laureate Paul Samuelson proved in 1942 that, in two dimensions, it is the only line expressible solely in terms of the ratios of standard deviations and the correlation coefficient which (1) fits the correct equation when the observations fall on a straight line, (2) exhibits scale invariance, and (3) exhibits invariance under interchange of variables.{{cite journal |last=Samuelson |first=Paul A. |year=1942 |title=A Note on Alternative Regressions |journal=Econometrica |doi=10.2307/1907024 |jstor=1907024 |volume=10 |issue=1 |pages=80–83}} This solution has been rediscovered in different disciplines and is variously known as standardised major axis (Ricker 1975, Warton et al., 2006),{{cite journal |last=Ricker |first=W. E. |year=1975 |title=A note concerning Professor Jolicoeur's Comments |journal=Journal of the Fisheries Research Board of Canada |doi=10.1139/f75-172 |volume=32 |issue=8 |pages=1494–1498}}{{cite journal |last1=Warton |first1=David I. |last2=Wright |first2=Ian J. |last3=Falster |first3=Daniel S. |last4=Westoby |first4=Mark |year=2006 |title=Bivariate line-fitting methods for allometry |journal=Biological Reviews |doi=10.1017/S1464793106007007 |volume=81 |issue=2 |pages=259–291|pmid=16573844 |citeseerx=10.1.1.461.9154 |s2cid=16462731 }} the reduced major axis, the geometric mean functional relationship (Draper and Smith, 1998),Draper, NR and Smith, H. Applied Regression Analysis, 3rd edition, pp. 92–96. 1998 least products regression, diagonal regression, line of organic correlation, and the least areas line (Tofallis, 2002).{{cite book |last=Tofallis |first=Chris |editor1-last=Van Huffel |editor1-first=Sabine |editor1-link= Sabine Van Huffel |editor2-last=Lemmerling |editor2-first=P. |year=2002 |title=Total Least Squares and Errors-in-Variables Modeling: Analysis, Algorithms and Applications |chapter=Model Fitting for Multiple Variables by Minimising the Geometric Mean Deviation |publisher=Kluwer Academic Publ. |location=Dordrecht |isbn=978-1402004766 |ssrn=1077322}}

Tofallis (2015, 2023){{ Cite SSRN |last=Tofallis|first=Chris|year=2015|title=Fitting Equations to Data with the Perfect Correlation Relationship |ssrn=2707593}}Tofallis, C. (2023). Fitting an Equation to Data Impartially. Mathematics, 11(18), 3957. https://ssrn.com/abstract=4556739 https://doi.org/10.3390/math11183957 has extended this approach to deal with multiple variables. The calculations are simpler than for total least squares as they only require knowledge of covariances, and can be computed using standard spreadsheet functions.

• Regression dilution
• Deming regression, a special case with two predictors and independent errors.
• Errors-in-variables model
• Gauss-Helmert model
• Linear regression
• Least squares
• Principal component analysis
• Principal component regression

{{reflist|group=note}}

{{Reflist}}

• I. Hnětynková, M. Plešinger, D. M. Sima, Z. Strakoš, and S. Van Huffel, The total least squares problem in AX ≈ B. A new classification with the relationship to the classical works. SIMAX vol. 32 issue 3 (2011), pp. 748–770. Available as a [https://www.sam.math.ethz.ch/sam_reports/reports_final/reports2010/2010-38_fp.pdf preprint].
• M. Plešinger, The Total Least Squares Problem and Reduction of Data in AX ≈ B. Doctoral Thesis, TU of Liberec and Institute of Computer Science, AS CR Prague, 2008. [https://www.fp.tul.cz/~plesinger/my_publications/doctoral_thesis/thesis.pdf Ph.D. Thesis]
• C. C. Paige, Z. Strakoš, Core problems in linear algebraic systems. SIAM J. Matrix Anal. Appl. 27, 2006, pp. 861–875. {{doi|10.1137/040616991}}
• S. Van Huffel and P. Lemmerling, Total Least Squares and Errors-in-Variables Modeling: Analysis, Algorithms and Applications. Dordrecht, The Netherlands: Kluwer Academic Publishers, 2002.
• S. Jo and S. W. Kim, Consistent normalized least mean square filtering with noisy data matrix. IEEE Trans. Signal Process., vol. 53, no. 6, pp. 2112–2123, Jun. 2005.
• R. D. DeGroat and E. M. Dowling, The data least squares problem and channel equalization. IEEE Trans. Signal Process., vol. 41, no. 1, pp. 407–411, Jan. 1993.
• S. Van Huffel and J. Vandewalle, The Total Least Squares Problems: Computational Aspects and Analysis. SIAM Publications, Philadelphia PA, 1991. {{doi|10.1137/1.9781611971002}}
• T. Abatzoglou and J. Mendel, Constrained total least squares, in Proc. IEEE Int. Conf. Acoust., Speech, Signal Process. (ICASSP’87), Apr. 1987, vol. 12, pp. 1485–1488.
• P. de Groen An introduction to total least squares, in Nieuw Archief voor Wiskunde, Vierde serie, deel 14, 1996, pp. 237–253 [https://arxiv.org/abs/math.RA/9805076/ arxiv.org].
• G. H. Golub and C. F. Van Loan, An analysis of the total least squares problem. SIAM J. on Numer. Anal., 17, 1980, pp. 883–893. {{doi|10.1137/0717073}}
• [http://www.mathpages.com/home/kmath110.htm Perpendicular Regression Of A Line] at MathPages
• A. R. Amiri-Simkooei and S. Jazaeri Weighted total least squares formulated by standard least squares theory, in Journal of Geodetic Science, 2 (2): 113–124, 2012 [http://engold.ui.ac.ir/~amiri/JGS_Amiri_Jazaeri_2012.pdf].

{{Least Squares and Regression Analysis}}

{{DEFAULTSORT:Total Least Squares}}

Category:Applied mathematics

Category:Curve fitting

Category:Least squares

Category:Regression models